Ricerche di Matematica manuscript No. (will be inserted by the editor) On attainability of optimal controls in coefficients for system of Hammerstein type with anisotropic p-Laplacian Tiziana Durante · Olha P. Kupenko · Rosanna Manzo 7 1 0 Received: date/Accepted: date 2 n a Abstract In this paper we consider an optimal control problem (OCP) for J thecoupledsystemofanonlinearmonotoneDirichletproblemwithanisotropic 3 p-Laplacianandmatrix-valuedL∞(Ω,RN×N)-controlsinitscoefficientsanda 2 nonlinear equation of Hammerstein type. Using the direct method in calculus ] ofvariations,we provethe existence of anoptimalcontrolin consideredprob- C lem and provide sensitivity analysis for a specific case of considered problem O with respect to two-parameter regularization. . h t Keywords Nonlinear elliptic equations Hammerstein equation control in a · · m coefficients p(x)-Laplacian approximation approach · · [ Mathematics Subject Classification (2000) 47H30 35B20 35M12 1 · · · 35J60 49J20 v · 9 0 T.Durante 6 Universit`adegliStudidiSalerno,DipartimentodiIngegneriadell’Informazione 6 edElettricaeMatematicaApplicata, 0 ViaGiovanni PaoloII,132,84084Fisciano(SA),Italy . 1 E-mail:[email protected] 0 O.Kupenko 7 NationalMiningUniversity,DepartmentofSystemAnalysisandControl, 1 Yavornitskyiav.,19,49005Dnipro,Ukraine, : NationalTechnical UniversityofUkraine“KievPolytechnical Institute”, v InstituteforAppliedandSystem Analysis, i X Peremogyav.,37,build.35,03056Kiev,Ukraine E-mail:kogut [email protected] r a R.Manzo Universit`adegliStudidiSalerno,DipartimentodiIngegneriadell’Informazione edElettricaeMatematicaApplicata, ViaGiovanni PaoloII,132,84084Fisciano(SA),Italy E-mail:[email protected] 2 T.Durante,O.P.Kupenko,R.Manzo 1 Introduction The aim of this paper is to prove the existence result for an optimal control problem (OCP) governed by the system of a homogeneous Dirichlet nonlin- ear elliptic boundary value problem, whose principle part is an anisotropic p-Laplace-like operator, and a nonlinear equation of Hammerstein type, and to provide sensitivity analysis for the specific case of considered optimiza- tion problem with respect to a two-parameter regularization. As controls we consider the symmetric matrix of anisotropy in the main part of the elliptic equation. We assume that admissible controls are measurable and uniformly bounded matrices of L∞(Ω;RN×N). Systemswithdistributedparametersandoptimalcontrolproblemsforsys- tems describedbyPDE,nonlinearintegralandordinarydifferentialequations havebeenwidelystudiedbymanyauthors(seeforexample[13,19,21,23,27]). However, systems which contain equations of different types and optimiza- tion problems associated with them are still less well understood. In general case including as well control and state constraints, such problems are rather complex and have no simple constructive solutions. The system, considered in the present paper, contains two equations: a nonlinear elliptic equation with the so-calledanisotropicp-Laplace operator with homogeneousDirichlet boundary conditions and a nonlinear equation of Hammerstein type, which nonlinearly depends on the solution of the first object. The optimal control problem we study here is to minimize the discrepancy between a given dis- tribution z Lp(Ω) and a solution of Hammerstein equation z = z(A,y), d choosing an a∈ppropriate matrix of coefficients A A , i.e. ad ∈ I(A,y,z)= z(x) z (x)2dx inf (1) d | − | −→ ZΩ subject to constrains z+BF(y,z)=0 in Ω, (2) div (A(x) y, y)RN (p−2)/2A(x) y =f in Ω, (3) − | ∇ ∇ | ∇ A A , y =0 on ∂Ω, (4) (cid:0) ad (cid:1) ∈ where B : Lq(Ω) Lp(Ω) is a positive linear operator, F : W1,p(Ω) → 0 × Lp(Ω) Lq(Ω) is a nonlinear operator, f L2(Ω) is a given distribu- tion, an→d a class of admissible controls A is∈a nonempty compact subset ad of L∞(Ω;RN(N2+1)). The interestto equationswhose principle partis ananisotropicp-Laplace- likeoperatorarisesfromvariousappliedcontextsrelatedtocompositemateri- als such as nonlinear dielectric composites, whose nonlinear behavior is mod- eledbytheso-calledpower-low(see,forinstance,[4,20]andreferencestherein). Itissufficienttosaythatanisotropicp-Laplacian∆ (A,y)hasprofoundback- p ground both in the theory of anisotropic and nonhomogeneous media and in Finsler or Minkowski geometry [31]. As a rule, the effect of anisotropy ap- pearsnaturallyinawideclassofgeometry—Finslergeometry.Atypicaland OptimalcontrolsforHammersteinsystemwithanisotropicp-Laplacian 3 important example of Finsler geometry is Minkowski geometry. In this case, anisotropicLaplacianis closelyrelatedtoaconvexhypersurfaceinRN,which iscalledtheWulffshape[30].SincethetopologyoftheWulffshapeessentially depends on the matrix of anisotropy A(x), it is reasonable to take such ma- trix as a control.Frommathematicalpointof view,the interestof anisotropic p-Laplacian lies on its nonlinearity and an effect of degeneracy, which turns out to be the major difference from the standard Laplacian on RN. Inpractice,theequationsofHammersteintypeappearasintegralorinteg- ro-differential equations. The class of integral equations is very important for theory and applications, since there are less restrictions on smoothness of the desired solutions involved in comparison to those for the solutions of differ- ential equations. It should be also mentioned here, that well posedness or uniqueness of the solutions is not typical for equations of Hammerstein type or optimization problems associated with such objects (see [3]). Indeed, this property requires rather strong assumptions on operators B and F, which is ratherrestrictiveinviewofnumerousapplications(see[25]).Thephysicalmo- tivationofoptimalcontrolproblemswhicharesimilarto thoseinvestigatedin the present paper is widely discussed in [3,26]. Using the direct method of the Calculus of Variations, we show in Section 4 that the optimal control problem (1)–(4) has a nonempty set of solutions provided the admissible controls A(x) are uniformly bounded in BV-norm, in spite of the fact that the corresponding quasilinear differential operator div (A y, y)RN p−22A y ,inprinciple,hasdegeneraciesas A12 y tends − | ∇ ∇ | ∇ | ∇ | p−2 to ze(cid:0)ro [1]. Moreover, when(cid:1)the term |(A∇y,∇y)RN| 2 is regarded as the coefficientofthe Laplaceoperator,wehavethe caseofunbounded coefficients (see [12,14]). In order to avoid degeneracy with respect to the control A(x), we assume that matrix A(x) has a uniformly bounded spectrum away from zero.As for the optimalcontrolproblems incoefficientsfor degenerateelliptic equations and variational inequalities, we can refer to [5,8,9,10,16,17,19]. A number of regularizations have been suggested in the literature. See [24] for a discussion for what has come to be known as (ε,p)-Laplace prob- lem, such as div((ε+ y 2)p−22) y. While the (ε,p)-Laplacian regularizes − |∇ | ∇ the degeneracy as the gradients tend to zero, the term y p−2, viewed again |∇ | as a coefficient, may grow large [6]. Therefore, following ideas of [7], for the specificcaseofconsideredoptimizationproblemweintroduceyetanotherreg- ularizationthat leads to a sequence of monotone and bounded approximation k(A21 y 2)of A12 y 2. As a result,forfixed parameterp [2, )andcon- F | ∇ | | ∇ | ∈ ∞ trolA(x),wearriveatatwo-parametervariationalproblemgovernedbyopera- tor div((ε+ k(A12 y 2))p−22)A yandatwo-parameterHammersteinequa- − F | ∇ | ∇ tion with non-linear kernel Fε,k(y,z)=(ε+ k(y2))p−22y+(ε+ k(z2))p−22z. F F Finally, we deal with a two-parameter family of optimal control problems in the coefficients for a system of elliptic boundary value problem and equation of Hammerstein type. We consequently provide the well-posedness analysis for the perturbed optimal control problems in Sections 5. In section 6, we showthatthe solutionsoftwo-parametricfamily ofperturbedoptimalcontrol 4 T.Durante,O.P.Kupenko,R.Manzo problems can be considered as appropriate approximations to optimal pairs for the original problem similar to (1)–(4). To the end, we note that the ap- proximation and regularization are not only considered to be useful for the mathematicalanalysis,butalsoforthe purposeofnumericalsimulations.The numerical analysis as well as the case of degenerating controls are subjects to future publications. 2 Notation and preliminaries Let Ω be a bounded open subset of RN (N 1) with a Lipschitz boundary. ≥ Let p be a real number such that 2 p < , and let q = p/(p 1) be ≤ ∞ − the conjugate of p. Let SN := RN(N2+1) be the set of all symmetric matrices A = [a ]N , (a = a R). We suppose that SN is endowed with the ij i,j=1 ij ji ∈ Euclidianscalar product A B =tr(AB)=a b andwith the corresponding ij ij · Euclidiannorm A SN =(A A)1/2.Wealsomakeuseoftheso-calledspectral norm A :=sukpkAξ : ξ· RN with ξ =1 of matrices A SN, which 2 k k | | ∈ | | ∈ is different from the Euclidean norm A SN. However, the relation A 2 A SN √N A 2(cid:8)holds true for all Ak SkN. (cid:9) k k ≤ k k ≤ k k ∈ Let L1(Ω)N(N2+1) = L1 Ω;SN be the space of integrable functions whose values are symmetric matrices. By BV(Ω;SN) we denote the space of all matrices in L1(Ω;SN) for(cid:0)which t(cid:1)he norm kAkBV(Ω;SN) =kAkL1(Ω;SN)+ |DA|=kAkL1(Ω;SN) ZΩ + sup a divϕdx : ϕ C1(Ω;RN), ϕ(x) 1 for x Ω ij ∈ 0 | |≤ ∈ 1≤iX≤j≤N nZΩ o (5) is finite. Weak Compactness Criterion in L1(Ω). Throughout the paper we will often use the concept of weak and strong convergence in L1(Ω). Let a { ε}ε>0 be abounded sequenceof functions inL1(Ω). We recallthat a is called { ε}ε>0 equi-integrableonΩ,ifforanyδ >0thereisaτ =τ(δ)suchthat a dx< Sk εk δ for every measurable subset S Ω of Lebesgue measure S < τ. Then the ⊂ | | R following assertions are equivalent for L1(Ω)-bounded sequences: (i) a sequence a is weakly compact in L1(Ω); { k}k∈N (ii) the sequence a is equi-integrable. { k}k∈N Lemma 1 (Lebesgue’s Theorem) If a sequence a L1(Ω) is equi- { k}k∈N ⊂ integrable and a a almost everywhere in Ω then a a in L1(Ω). k k → → Lemma 2 ([28]) If a sequence ϕk k∈N is bounded in L1(Ω), ϕk 0 a.e. in { } → Ω and gk k∈N is equi-integrable, then ϕk gk 0 strongly in L1(Ω). { } · → OptimalcontrolsforHammersteinsystemwithanisotropicp-Laplacian 5 Lemma 3 ([28]) Let B (x,ξ) and B(x,ξ) be Caratheodory vector functions n acting from Ω R to R. These vector functions are assumed to satisfy the × monotonicity and pointwise convergence conditions (B (x,ξ) B (x,η))(ξ η) 0, B (x,0) 0, n n n − − ≥ ≡ B (x,ξ) c (ξ )< ; lim B (x,ξ)=B(x,ξ), n 0 n | |≤ | | ∞ n→∞ for a.e. x Ω and every ξ R. If v ⇀v in Lp(Ω), B (x,v )⇀z in Lq(Ω), n n n ∈ ∈ then liminf B (x,v ),v z,v , (6) n n n Lq(Ω);Lp(Ω) Lq(Ω);Lp(Ω) n→∞ h i ≥h i and in the case of equality in (6), we have z =B(x,v). Admissible controls. Let ξ , ξ be given elements of L∞(Ω) BV(Ω) sat- 1 2 ∩ isfying the conditions 0<α ξ (x) ξ (x) a.e. in Ω, (7) 1 2 ≤ ≤ where α is a given positive value. We define the class of admissible controls A as follows ad A = A L∞(Ω;SN) ξ12|η|2 ≤(η,Aη)RN ≤ξ22|η|2a.e.inΩ, ∀η ∈RN, ad ( ∈ (cid:12) A12 BV(Ω;SN), DA21 γ, ) (cid:12) ∈ Ω| |≤ (cid:12) (8) (cid:12) R where γ >0 is a given con(cid:12)stant. In view of estimate A21(x) SN √N A21(x) 2 √Nξ2(x) a.e. in Ω, k k ≤ k k ≤ it is clear that A is a nonempty convex subset of L∞(Ω;SN). ad Anisotropic Laplace operator. Let us consider now the nonlinear operator (A,y):L∞(Ω;SN) W1,p(Ω) W−1,q(Ω) defined as A × 0 → p−2 (A,y)= div (A y, y)RN 2 A y A − | ∇ ∇ | ∇ or via the pairing (cid:0) (cid:1) p−2 hA(A,y),viW−1,q(Ω);W01,p(Ω) :=ZΩ|(A∇y,∇y)RN| 2 (A∇y,∇v)RN dx = |A12∇y|p−2(A∇y,∇v)RN dx, ∀v ∈W01,p(Ω). (9) ZΩ Definition 1 We say that a function y = y(A,f) is a weak solution (in the sense of Minty) to boundary value problem p−2 div (A y, y)RN 2 A y =f in Ω, (10) − | ∇ ∇ | ∇ (cid:0) A∈Aad, y ∈W01,p(Ω(cid:1)), (11) 6 T.Durante,O.P.Kupenko,R.Manzo for a fixed control A A and given function f L2(Ω) if the inequality ad ∈ ∈ |A1/2∇ϕ|pR−N2(A∇ϕ,∇ϕ−∇y)RN dx≥ f(ϕ−y)dx (12) ZΩ ZΩ holds for any ϕ C∞(Ω). ∈ 0 Remark 1 Anotherdefinitionoftheweaksolutiontotheconsideredboundary value problem appears more natural: |A1/2∇y|pR−N2(A∇y,∇ϕ)RN dx= fϕdx∀ϕ∈C0∞(Ω). ZΩ ZΩ However,bothconceptsfortheweaksolutionscoincide(see,forinstance,[29]). Let us show that for each A A operator () = (A, ) : W1,p(Ω) ∈ ad A · A · 0 → W−1,q(Ω)isstrictlymonotone,coerciveandsemi-continuous,wheretheabove mentioned properties have respectively the following meaning: y v,y v 0, y,v W1,p(Ω); (13) hA −A − iW−1,q(Ω);W01,p(Ω) ≥ ∀ ∈ 0 y v,y v =0 = y =v; (14) hA −A − iW−1,q(Ω);W01,p(Ω) ⇒ y,y + provided y ; (15) hA iW−1,q(Ω);W01,p(Ω) → ∞ k kW01,p(Ω)→∞ R t (y+tv),w is continuous y,v,w W1,p(Ω). ∋ 7→hA iW−1,q(Ω);W01,p(Ω) ∀ ∈ 0 (16) Indeed, the right-hand side of (9) is continuous with respect to v W1,p(Ω) ∈ 0 and, therefore, represents an element of W−1,q(Ω) because p−1 1 |A21∇y|p−2(A∇y,∇v)RN dx≤ |A21∇y|pdx p |A12∇v|pdx p ZΩ (cid:18)ZΩ (cid:19) (cid:18)ZΩ (cid:19) ξ p y p−1 v = ξ p y p−1 v ≤k 2kL∞(Ω)k∇ kLp(Ω)Nk∇ kLp(Ω)N k 2kL∞(Ω)k kW01,p(Ω)k kW01,p(Ω) i(nwgeafrpopmlythheerectohnediHtio¨olndeAr’sineAqua)l.ityHeanncde,esftoimr aetaech|AA12∇ϕ|Ap ≤ξt2ph|∇e ϕop|percaotmor- ad ad (A, ) : W1,p(Ω) W−1,q∈(Ω) is bounded. The coercivi∈ty property of we A · 0 → A get immediately, since y,y αp y p . hA iW−1,q(Ω);W01,p(Ω) ≥ k kW01,p(Ω) As for the proof ofthe strictmonotonicity and semicontinuity of the operator ,wereferforthedetailsto[22,24]).Then,bywellknownexistenceresultsfor A non-linearellipticequationswithcoercive,semi-continuous,strictlymonotone operators, the Dirichlet boundary value problem (10)–(11) admits a unique weak solution for every fixed control matrix A A and every distribution ad ∈ f L2(Ω). ∈ On equations of Hammerstein type. Let Y and Z be Banach spaces, let Y Y be an arbitrary bounded set, and let Z∗ be the dual space to Z. To 0 ⊂ beginwithwerecallsomeusefulpropertiesofnon-linearoperators,concerning the solvability problem for Hammerstein type equations and systems. OptimalcontrolsforHammersteinsystemwithanisotropicp-Laplacian 7 Definition 2 We say that the operator G : D(G) Z Z∗ is radially ⊂ → continuous if for any z ,z X there exists ε>0 such that z +τz D(G) 1 2 1 2 ∈ ∈ for all τ [0,ε] and a real-valued function [0,ε] τ G(z +τz ),z 1 2 2 Z∗;Z ∈ ∋ → h i is continuous. Definition 3 An operator G:Y Z Z∗ is said to have a uniformly semi- × → bounded variation (u.s.b.v.) if for any bounded set Y Y and any elements 0 ⊂ z ,z D(G) such that z R, i=1,2, the following inequality 1 2 i Z ∈ k k ≤ G(y,z ) G(y,z ),z z inf C (R; z z ) (17) 1 2 1 2 Z∗;Z y 1 2 Z h − − i ≥−y∈Y0 k| − k| holds true provided the function C : R R R is continuous for each y + + × → 1 element y Y , and C (r,t) 0 as t 0, r > 0. Here, is a 0 y Z ∈ t → → ∀ k|·k| seminorm on Z such that is compact with respect to the norm . Z Z k|·k| k·k It is worth to note that Definition 3 gives in fact a certain generalization of the classical monotonicity property. Indeed, if C (ρ,r) 0, then (17) im- y ≡ plies the monotonicity property for the operatorG with respect to the second argument. Remark 2 EachoperatorG:Y Z Z∗ withu.s.b.v.possessesthefollowing × → property (see for comparisonRemark 1.1.2 in [3]): if a set K Z is such that ⊂ z k and G(y,z),z k for all z K and y Y , then there Z 1 Z∗;Z 2 0 k k ≤ h i ≤ ∈ ∈ exists a constant C >0 such that G(y,z) C, z K and y Y . Z∗ 0 k k ≤ ∀ ∈ ∀ ∈ Let B : Z∗ Z and F : Y Z Z∗ be given operators such that the → × → mapping Z∗ z∗ B(z∗) Z is linear. Let g Z be a given distribution. ∋ 7→ ∈ ∈ Then a typical operator equation of Hammerstein type can be represented as follows z+BF(y,z)=g. (18) The following existence result is well-known (see [3, Theorem 1.2.1]). Theorem 1 Let B : Z∗ Z be a linear continuous positive operator such → that there exists a right inverse operator B−1 :Z Z∗. Let F :Y Z Z∗ r → × → be an operator with u.s.b.v such that F(y, ) : Z Z∗ is radially continuous · → for each y Y and the following inequality holds true 0 ∈ F(y,z) B−1g,z 0 if only z >λ>0, λ=const. h − r iZ∗;Z ≥ k kZ Then the set (y)= z Z : z+BF(y,z)=g in the sense of distributions H { ∈ } is non-empty and weakly compact for every fixed y Y and g Z. 0 ∈ ∈ In what follows, we set Y =W1,p(Ω), Z =Lp(Ω), and Z∗ =Lq(Ω). 0 8 T.Durante,O.P.Kupenko,R.Manzo 3 Setting of the optimal control problem Let us consider the following optimal control problem: Minimize I(A,y,z)= z(x) z (x)2dx , (19) d | − | n ZΩ o subject to the constraints |A1/2∇y|p−2(A∇y,∇ϕ)RN dx= fϕdx, ∀ϕ∈W01,p(Ω), (20) ZΩ ZΩ A A , y W1,p(Ω), (21) ∈ ad ∈ 0 zφdx+ BF(y,z)φdx=0, φ Lq(Ω), (22) ∀ ∈ ZΩ ZΩ wheref L2(Ω) andz Lp(Ω) aregivendistributions, B :Lq(Ω) Lp(Ω) d is a linea∈r operator, F :W∈1,p(Ω) Lp(Ω) Lq(Ω) is a non-linear o→perator. Letus denote by Ξ L0∞(Ω;S×N) W1→,p(Ω) Lp(Ω) the setof alladmis- ⊂ × 0 × sible triplets to the optimal control problem (19)–(22). Hereinafter we suppose that the space L∞(Ω;SN) W1,p(Ω) Lp(Ω) is × 0 × endowedwiththenormk(A,y,z)kL∞(Ω;SN)×W01,p(Ω)×Lp(Ω) :=kA12kBV(Ω;SN)+ y + z . k kW01,p(Ω) k kLp(Ω) Remark 3 Werecallthatasequence f ∞ convergesweakly-∗tof inBV(Ω) { k}k=1 if and only if the two following conditions hold (see [2]): f f strongly in k L1(Ω) and Df ⇀∗ Df weakly∗ in the space of Radon measu→res (Ω;RN). k M Moreover, if f ∞ BV(Ω) converges strongly to some f in L1(Ω) and { k}k=1 ⊂ satisfies sup Df <+ , then (see, for instance, [2]) k∈N Ω| k| ∞ R (i) f BV(Ω) and Df liminf Df ; k ∈ ZΩ| |≤ k→∞ ZΩ| | (23) ∗ (ii) f ⇀f in BV(Ω). k Alsowerecall,thatuniformlyboundedsetsinBV-normarerelativelycompact in L1(Ω). Definition 4 We say that a sequence of triplets (Ak,yk,zk) k∈N from the space L∞(Ω;SN) W1,p(Ω) Lp(Ω) τ-converges{to a triplet}(A ,y ,z ) if × 0 × 0 0 0 A21 ⇀∗ A12 in BV(Ω;SN), y ⇀y in W1,p(Ω) and z ⇀z in Lp(Ω). k 0 k 0 0 k 0 Further we use the following auxiliary results. Proposition 1 For each A A and every f L2(Ω), a weak solution y to ad ∈ ∈ variational problem (20)–(21) satisfies the estimate q y α−q f p . (24) k kW01,p(Ω) ≤ k kW−1,q(Ω) OptimalcontrolsforHammersteinsystemwithanisotropicp-Laplacian 9 Proof The estimate (24) immediately follows from the following relations αpkykpW01,p(Ω) ≤ZΩ|A12∇y|pdx=hA(A,y),yiW−1,q(Ω);W01,p(Ω) = f,y f y . h iW−1,q(Ω);W01,p(Ω) ≤k kW−1,q(Ω)k kW01,p(Ω) Lemma 4 Let (A ,y ) L∞(Ω;SN) W1,p(Ω) be a sequenceof pairs k k ∈ × 0 k∈N such that Ak ∈nAad ∀k ∈ N, Ak12 ⇀∗ A12 in BVo(Ω;SN), and yk ⇀ y in W1,p(Ω). Then 0 p−2 kl→im∞ZΩ|(∇ϕ,Ak∇ϕ)RN | 2 (∇yk,Ak∇ϕ)RN dx = |(∇ϕ,A∇ϕ)RN |p−22 (∇y,A∇ϕ)RN dx, ∀ϕ∈C0∞(Ω). (25) ZΩ Proof Since Ak21 →A12 in L1(Ω;SN) and {Ak}k∈N is bounded in L∞(Ω;SN), by Lebesgue’s Theorem we get that Ak12 → A21 strongly in Lr(Ω;SN) for every 1 ≤ r < +∞. Hence, Ak21∇ϕ → A12∇ϕ strongly in Lp(Ω)N for every ϕ C∞(Ω). Therefore, ∈ 0 |Ak12∇ϕ|p−2Ak21∇ϕ→|A21∇ϕ|p−2A21∇ϕ in Lq(Ω)N, ∀ϕ∈C0∞(Ω). (26) Moreover,sinceAk12∇ψ →A12∇ψstronglyinLq(Ω)N foreveryψ ∈C0∞(Ω) and y ⇀ y in Lp(Ω)N, it follows that k ∇ ∇ 1 1 A2 y , ψ dx= y ,A2 ψ dx ZΩ(cid:16) k∇ k ∇ (cid:17)RN ZΩ(cid:16)∇ k k∇ (cid:17)RN →ZΩ(cid:16)∇y,A21∇ψ(cid:17)RN dx=ZΩ(cid:16)A12∇y,∇ψ(cid:17)RN dx, ∀ψ ∈C0∞(Ω) (27) as a product of weakly and strongly convergent sequences in Lp(Ω)N and Lq(Ω)N, respectively. Using the fact that 1 sup A2 y ξ sup y <+ , k∈Nk k∇ kkLp(Ω)N ≤k 2kL∞(Ω)k∈Nk∇ kkLp(Ω)N ∞ we finally get from (27) Ak12∇yk ⇀A21∇y in Lp(Ω)N. (28) Thus, to complete the proof it remains to note that p−2 |(∇ϕ,Ak∇ϕ)RN | 2 (∇yk,Ak∇ϕ)RN dx ZΩ = A12 ϕp−2A21 ϕ,A21 y dx ZΩ(cid:16)| k∇ | k∇ k∇ k(cid:17)RN and apply the properties (26) and (28). 10 T.Durante,O.P.Kupenko,R.Manzo Thefollowingresultconcernstheregularityoftheoptimalcontrolproblem (19)–(22). Proposition 2 Let B : Lq(Ω) Lp(Ω) and F : W1,p(Ω) Lp(Ω) Lq(Ω) → 0 × → be operators satisfying all conditions of Theorem 1. Then the set Ξ = (A,y,z) L∞(Ω;SN) W1,p(Ω) Lp(Ω): ∈ × 0 × (cid:8) (A,y)=f, z+BF(y,z)=0 A is nonempty for every f L2(Ω). (cid:9) ∈ Proof Let A A be an arbitrary admissible control. Then for a given ad ∈ f L2(Ω), the Dirichlet boundary problem (20)–(21) admits a unique so- lut∈iony =y(A,f) W1,p(Ω)whichsatisfiesthe estimate(24).Itremainsto A ∈ 0 remark that the corresponding Hammerstein equation z+BF(y ,z)=0 has A a nonempty set of solutions (y ) by Theorem 1. A H 4 Existence of optimal solutions Thefollowingresultiscrucialforourconsiderationanditstatesthefact,that thesetofadmissibletripletstotheoptimalcontrolproblem(19)–(22)isclosed with respect to τ-topology of the space L∞(Ω;SN) W1,p(Ω) Lp(Ω). × 0 × Theorem 2 Assume the following conditions hold: – The operators B : Lq(Ω) Lp(Ω) and F : W1,p(Ω) Lp(Ω) Lq(Ω) → 0 × → satisfy conditions of Theorem 1; – The operator F(,z):W1,p(Ω) Lq(Ω) is compact in the following sense: if y ⇀y weakly· in W10,p(Ω), t→hen F(y ,z) F(y ,z)strongly in Lq(Ω). k 0 0 k → 0 Then for every f L2(Ω) the set Ξ is sequentially τ-closed, i.e. if a sequence (Ak,yk,zk) Ξ∈k∈N τ-converges to a triplet (A0,y0,z0) L∞(Ω;SN) {W1,p(Ω) Lp∈(Ω),}then A A , y = y(A ), z (y )∈, and, therefore×, 0 × 0 ∈ ad 0 0 0 ∈ H 0 (A ,y ,z ) Ξ. 0 0 0 ∈ Proof Let (Ak,yk,zk) k∈N Ξ be any τ-convergent sequence of admissible { } ⊂ triplets to the optimal control problem (19)–(22), and let (A ,y ,z ) be its 0 0 0 τ-limit in the sense of Definition 4. We divide the rest of the proof onto two steps. Step 1. On this step we show that A A and y = y(A ). As follows 0 ad 0 0 ∈ from Definition 4 and Remark 3, we have A12 A12 in L1(Ω;SN), y ⇀y in W1,p(Ω), (29) k → 0 k 0 0 1 1 A2 A2 almost everywhere in Ω, (30) k → 0 1 1 DA2 liminf DA2 γ. (31) ZΩ| 0|≤ k→∞ ZΩ| k|≤